SUPERLUMINOUS SUPERNOVA SN 2015bn IN THE NEBULAR PHASE: EVIDENCE FOR THE ENGINE-POWERED EXPLOSION OF A STRIPPED MASSIVE STAR

Superluminous supernovae (SLSNe) are the brightest explosions in the universe at optical wavelengths (see Gal-Yam 2012), but long evaded detection due to their rarity and preference for low-metallicity dwarf galaxies (e.g., Chen et al. 2013; Lunnan et al. 2014; Leloudas et al. 2015). Since they were first recognized (Quimby et al. 2011), the number of known SLSNe—almost invariably discovered by untargeted transient surveys—has grown to several tens (Nicholl et al. 2015).

The spectra of Type I SLSNe show neither hydrogen nor low-velocity lines to clearly indicate interaction with circumstellar medium (CSM). A few events have exhibited hydrogen emission after several hundred days, but this is not directly related to their power source at peak (Yan et al. 2015). Early spectra are instead dominated by O ii absorption and blue continua indicating hot photospheres. After sufficient cooling, they evolve to resemble normal and broad-lined SNe Ic of more typical luminosity (Pastorello et al. 2010).

An appealing model to power SLSNe I is a "central engine" (such as a millisecond magnetar; Shklovskii 1976; Duncan & Thompson 1992; Kasen & Bildsten 2010; Woosley 2010) with a timescale similar to the diffusion time, thus thermalizing the input without significant adiabatic dilution (Metzger et al. 2015). Evidence for this scenario includes light curve fitting (Inserra et al. 2013; Nicholl et al. 2013), similar host galaxies to long gamma-ray bursts (LGRBs; Lunnan et al. 2014; Perley et al. 2016), and large kinetic energies (Nicholl et al. 2015). Recently, the ultra-long GRB 111209A was associated with a very bright (though not quite superluminous) supernova, SN 2011kl (Greiner et al. 2015).

A crucial test for any model is provided by observations hundreds of days after explosion when the ejecta are largely transparent, offering direct constraints on the internal conditions and distribution of material. Only two SLSNe have spectra later than 300 days after maximum light: SN 2007bi (Gal-Yam et al. 2009) and PTF12dam (Chen et al. 2015). These have been of low signal-to-noise or strongly contaminated by the host (Jerkstrand et al. 2016a). In this Letter, we present deep photometric and spectroscopic observations of SN 2015bn—the best-observed SLSN to date (Nicholl et al. 2016)—at 250 to 400 days after peak¹³ , and show that the nebular spectrum connects SLSNe to other energetic and potentially engine-powered explosions spanning a wide range in luminosity.

2.1. New Data

2.2. Photometry

The light curves of SN 2015bn are shown in Figure 1. Photometry spanning the full range of u to K (made possible by PESSTO's dense coverage with both EFOSC2 and SOFI) is unprecedented for an SLSN at such late times (Chen et al. 2015 presented gri out to 400 days for PTF12dam). The luminosity at all wavelengths shows a slow decline, with little color evolution. We also plot the bolometric light curve, including K-corrections from spectra. The temperature and radius of a blackbody fit follows smoothly from those presented by Nicholl et al. (2016), but in fact most of the observed radiation is not thermal, and thus the late-time estimates of these properties are not reliable (Section 2.3).

Zoom In Zoom Out Reset image size

The previously published light curve out to +250 days appears roughly compatible with the decay rate of radioactive ⁵⁶Co (although we were unable to find a convincing radioactive decay model; Nicholl et al. 2016). However, our extended light curve shows that SN 2015bn fades significantly faster than fully trapped ⁵⁶Co decay. Linear fits from +60 to +400 days indicate a fading rate of 0.0143 mag day⁻¹, compared to 0.0098 for ⁵⁶Co. The decline is even steeper (0.017 mag day⁻¹) between 250 and 400 days. However, the steepening could be interpreted as another undulation in the light curve, such as those around maximum light. The decline rate is actually similar to that of GRB-SN 1998bw (Patat et al. 2001), but SN 2015bn is brighter by a factor of ∼150 at the same phase from explosion. The ⁵⁶Ni mass in SN 1998bw has been estimated as ≈0.3–0.7 M_⊙ (e.g., Sollerman et al. 2000; Mazzali et al. 2001); therefore, the luminosity of SN 2015bn would require at least a few tens of solar masses of ⁵⁶Ni to power the late-time light curve. However, a precise measurement is difficult as long as the light curve fades faster than ⁵⁶Co-decay. Applying Arnett's rule (Arnett 1982) to the light curve peaks and assuming rise times of 80 days and 15 days would give nickel masses of 32 M_⊙ and 0.4 M_⊙ for SN 2015bn and SN 1998bw, respectively. Given the long rise time of SN 2015bn, we have included ⁵⁶Co-decay in this calculation.

2.3. Spectroscopy

Our late-time spectroscopic series is shown in Figure 2, along with some early spectra from Nicholl et al. (2016). The spectrum at +243 days shows some emission lines—[Ca ii] λλ7291, 7324, Ca ii H&K, and possibly Mg i] λ4571—but still retains a photospheric component, with a blue continuum and absorption features. In our next spectrum, 20 days later, the Mg i] line strengthens significantly and O i starts to appear in emission, where we had previously detected it in absorption or P Cygni profiles. By +295 days, we see clear oxygen emission lines: [O i] λλ6300, 6364, [O i] λ5577, and O i λ7774, as well as Na i D and broad features around 5000 Å normally interpreted as magnesium and iron (e.g., Mazzali et al. 2001; Jerkstrand et al. 2016b). In particular, [O i] λλ6300, 6364 increases in strength beyond +300 days and is by far the strongest line by our final spectrum at +392 days. That the ejecta take ∼400 days to become nebular suggests both a large mass and a persistent source of ionizing radiation to maintain a photosphere to such late times. In contrast to the SLSNe studied by Yan et al. (2015), we see no hydrogen or interaction signatures. The [Ca ii]/[O i] flux ratio is about 0.5 after ≳300 days, which is similar to most SNe Ic (Elmhamdi 2011) but very different from massive pair-instability SN models (Dessart et al. 2013; Jerkstrand et al. 2016a). We detect features consistent with [O iii], though not as strong as those recently seen in SLSNe PS1-14bj and LSQ14an (C. Inserra et al. 2016, in preparation; Jerkstrand et al. 2016b; Lunnan et al. 2016).

Zoom In Zoom Out Reset image size

The [O i] λλ6300, 6364 line is often used as a diagnostic of asphericity (Taubenberger et al. 2009). Many H-poor SNe show a double-peaked line profile suggesting a bipolar explosion viewed at a large angle from the dominant axis (Mazzali et al. 2005; Maeda et al. 2008). Inserra et al. (2016) recently showed from spectropolarimetry that SN 2015bn displays a dominant axis. If it is a highly asymmetric or jet-driven explosion, the single-peaked profile we observe would point to a modest inclination.

The relatively narrower width of the O i λ7774 line suggests that the emission comes from two distinct velocity regions (Milisavljevic et al. 2013); its lower velocity could indicate that it is arising deeper in the ejecta. This density-sensitive recombination line (Maurer & Mazzali 2010; see also Section 3) is not typically seen with such strength in normal SNe Ic, but has been detected in a number of energetic SNe Ic as well as SLSN 2007bi (Gal-Yam et al. 2009; Milisavljevic et al. 2013).

In fact, our nebular spectrum exhibits striking similarity to energetic SNe Ic¹⁵ such as SN 1998bw (Patat et al. 2001), SN 1997dq (Taubenberger et al. 2009), and SN 2012au (Milisavljevic et al. 2013). These events are often termed "hypernovae" as their inferred kinetic energy is ≳10⁵² erg: an order of magnitude larger than in normal neutrino-driven SNe, thus seeming to require an additional engine (Iwamoto et al. 1998). The observational link between some hypernovae and LGRBs, demonstrated spectacularly by SN 1998bw (Galama et al. 1998), confirms this engine as most likely a rapidly rotating compact object: either an accreting black hole "collapsar" (MacFadyen & Woosley 1999) or a millisecond magnetar (Duncan & Thompson 1992). The extraordinary similarity in nebular-phase spectra (probing the conditions of the innermost ejecta from the stellar interior) demonstrates that SLSNe and hypernovae have similar conditions in their cores, This could indicate that their progenitors or explosion mechanisms are related, consistent with both classes occurring in similar host environments (Lunnan et al. 2014; Perley et al. 2016).

Given this clear link between SLSNe and hypernovae/GRB-SNe, we look to build a consistent picture of SN 2015bn within the central-engine framework. Independent evidence for this link comes from spectropolarimetry (Inserra et al. 2016), which shows axisymmetry similar to GRB-SNe. While black hole accretion has also been proposed as a viable engine for SLSNe (Dexter & Kasen 2013), magnetar-powered models are likely more applicable here due to the long engine timescale required by the observations.

Although the progenitors and explosion mechanism may be similar, it seems that a different process supplies the luminosity of SN 2015bn compared to the hypernovae (which seem to be heated by ⁵⁶Ni, e.g., Cano et al. 2016). In Section 2.2, we saw that the larger luminosity of SN 2015bn compared to SN 1998bw would require ≳30 M_⊙ of ⁵⁶Ni, but the spectroscopic similarity demonstrates that SN 2015bn cannot have an enormously larger ⁵⁶Ni fraction than the hypernovae. While 30 M_⊙ of ⁵⁶Ni could be produced in a pair-instability SN, our spectra do not resemble pair-instability models (Jerkstrand et al. 2016a), nor do we see the [Fe iii] lines that dominate SNe Ia in the blue. This ⁵⁶Ni-mass is also comparable to our largest estimates of the total ejecta mass in SN 2015bn (see below). With no strong signatures of CSM interaction, it seems that the engine itself most likely supplies the luminosity.

SN 2015bn does appear to be slightly brighter in the blue than SNe 1997dq and 2012au. This could point to a larger iron line luminosity, but we argue that it is more likely due to residual continuum flux. In Figure 3, we show that the spectroscopic evolution from ∼250–400 days can be well-reproduced by a superposition of the nebular spectrum of SN 1997dq and a blue continuum (with the exception of [O i] at earlier epochs). The models we show assume a blackbody form for the continuum; we use a temperature ≈12,000 K at all phases. However, this is simply intended to illustrate the blue color, rather than assert that the continuum is necessarily thermal. By adding continuum rather than [Fe ii] alone, we find a good match to magnesium and calcium lines. This further removes any need for an unusually large iron mass (see also Nicholl et al. 2013, supplementary information).

Zoom In Zoom Out Reset image size

The continuum component fades over time as the emission lines increase in strength and has little effect by the final spectrum. It is consistent with photospheric recession toward low mass-coordinate. An alternative explanation is that we are seeing leakage of radiation from the engine, although this is unlikely as most of this power should be emitted at much higher frequency (Metzger et al. 2014). Finally, we consider the possibility that the continuum is powered by late-time circumstellar interaction. However, the spectrum is nearly identical to SN 2012au, for which radio observations revealed a low-density CSM and only weak mass loss in the years before explosion (Kamble et al. 2014). Deep radio limits on SN 2015bn at +240 days from Nicholl et al. (2016) did not find evidence for interaction (but could not definitively exclude it).

That some of the power input appears to be thermalized even at such late stages suggests a high-density inner region (Mazzali et al. 2004). A defining prediction of magnetar-powered models is the formation of a constant-density shell due to the central overpressure from the magnetized wind (Kasen & Bildsten 2010; Woosley 2010; Metzger et al. 2014). This shell could perhaps also help to account for the relative strength (and width) of the O i λ7774 recombination line in SLSNe/hypernovae, if its electron density is ≳10⁸ cm⁻³ at this phase (Mazzali et al. 2004; Nicholl et al. 2013; Jerkstrand et al. 2016b). Normal SNe Ic without an engine—or GRB-SNe, where the engine punches right through the stellar envelope rather than forming a shell—do not exhibit a strong line here.

Taking Equation (7) from Kasen & Bildsten (2010) for the shell velocity, ${v}_{\mathrm{sh}}\approx {(2{E}_{\mathrm{sn}}/{M}_{\mathrm{ej}})}^{1/2}{(1+{E}_{\mathrm{mag}}/{E}_{\mathrm{sn}})}^{(1/2)}$, with the deposited magnetar energy E_mag and ejected mass M_ej from our model fits (Figure 4; see also Nicholl et al. 2016) we find v_sh ≳ 5000 km s⁻¹, consistent with the width of the permitted O i emission. However, we note that in simple spherical symmetry a dense shell results in a flat-topped line profile, in tension with the sharply peaked line we observe here. Multi-dimensional effects should be explored, but this is beyond the scope of our study. As this is a recombination line, a further requirement is that some oxygen remains ionized in the inner ejecta (in contrast to the broad [O i], which is thermally excited and forms further out in the fast, neutral ejecta). The hard radiation field of the engine provides a natural explanation for persistent ionization.

Zoom In Zoom Out Reset image size

Using the [O i] λλ6300, 6364 flux in our latest 3 spectra (≈3 × 10⁻¹⁵ erg s⁻¹ cm⁻²), we can estimate the oxygen mass assuming the line is optically thin (Uomoto 1986):

$$\begin{eqnarray}&&{M}_{{\rm{O}}}\,=\,{10}^{8}\,{D}^{2}\,{F}_{[{\rm{O}}{\rm{I}}]}\,\exp (\mathrm{22,800}\,{\rm{K}}/T),\end{eqnarray} \tag{ 1 }$$

where D is distance to the SN (529 Mpc), and T its temperature, which we estimate following Jerkstrand et al. (2014): the observed ratio [O i] λ5577/[O i] λλ6300, 6364 = 0.3 gives T ≈ 4900 K. This corresponds to ≈9 M_⊙. We find the same result using Equation (3) from Jerkstrand et al. Mazzali et al. (2001) found good agreement between oxygen masses derived with Equation (1) and from synthetic spectra and that oxygen constituted 60%–70% of the ejected mass, giving us a total mass M_ej ≈ 14 M_⊙. This should be treated with caution due to uncertain physical conditions and an exponential temperature dependence: taking the plausible range 4000–6000 K based on nebular models of hypernovae (Mazzali et al. 2001), the allowed range is M_O = 4–25 M_⊙, giving M_ej = 6–40 M_⊙. Jerkstrand et al. (2016b) have calculated detailed spectroscopic models for SN 2015bn at +300 days and find that a similar ejecta mass is required. Interestingly, Mazzali et al. (2016) derived similar masses for other SLSNe using spectroscopic models of the photospheric phase; thus, a consistent picture is emerging. With our best mass estimate and the observed line velocities (≈7500 km s⁻¹), we infer a kinetic energy ∼5 × 10⁵¹ erg—higher than that in canonical core-collapse SNe, and closer to hypernovae.

We fit the light curve of SN 2015bn with a magnetar-powered model in Figure 4. We show fits up to +250 days after peak and to the full light curve at +400 days; only small changes in parameters are required (see the caption). The parameters are very similar to those from Nicholl et al. (2016). With a magnetic field in the range 10¹³–10¹⁴ G, we do not expect the SN to be accompanied by a GRB (Metzger et al. 2015), consistent with the radio non-detection from Nicholl et al. (2016). The ejecta mass range, M_ej = 7–12 M_⊙, was calculated assuming an optical opacity κ_opt = 0.2 cm² g⁻¹. This is appropriate for singly ionized matter, which is a good approximation around maximum light (oxygen-dominated ejecta at T ≳ 10,000 K). However, after the ejecta cool, the opacity may fall more in line with normal SNe Ic, usually taken to be 0.05–0.1 cm² g⁻¹ (e.g., see Inserra et al. 2013). This would increase the derived diffusion mass by a factor of 1.6–2.5. Thus, we estimate a total plausible range of M_ej = 7–30 M_⊙, similar to estimates from spectroscopy.

The basic form of the magnetar model eventually follows a t⁻² decline, which is shallower than the apparently exponential decline we observe. This discrepancy may be a sign of time-variable opacity to the X-rays/γ-rays from the magnetar (this is distinct from the optical opacity governing the subsequent diffusion of thermalized photons). We demonstrate this with representative fits over the range of γ-ray opacities studied by Kotera et al. (2013), from κ_γ = 0.01 to 0.1 cm² g⁻¹, with lower κ_γ giving a steeper decline. Variable κ_γ could result from changes in ionization state and may be related to the other bumps and wiggles in the light curve (Nicholl et al. 2016).

We compare the putative leakage of the magnetar radiation (i.e., the difference between model and data as a function of time) to the fraction of the total luminosity in emission lines (represented by the strongest [O i] line). The absolute line flux decreases with time, but the emission relative to the pseudo-continuum gives a measure of just how nebular are the ejecta (see also Figure 3). A consistent picture emerges: as the ejecta become more nebular (the optical depth decreases), more of the magnetar radiation escapes without being thermalized, giving a larger deficit in the light curve compared to the model (which assumes an optically thick photosphere).

The deficit at 350–400 days is ≈1.4 × 10⁴² erg s⁻¹. If this energy escapes in the X-rays (Metzger et al. 2014), the flux reaching us would be F_X ≈ 4×10⁻¹⁴ erg s⁻¹ cm⁻² assuming no intervening absorption. Detecting this would provide independent confirmation of the magnetar model. Preliminary analysis of ∼15 ks of Swift-XRT data collected between 2016 July 3 and 19 puts a limit of F_X < 3.6 × 10⁻¹⁴ erg s⁻¹ cm⁻² on the X-ray flux (3 σ, 0.3–10 keV, unabsorbed power-law model with index γ = 2), and thus rules out intrinsic absorption columns N_H < 10²¹ cm⁻² if the excess of energy is effectively leaking out in the 0.3–10 keV energy band. Other observations are planned and will provide tighter constraints to the model. We note that X-rays have been detected in one previous SLSN by Levan et al. (2013). It is also possible that some of the missing flux could be accounted for with mid-/far-IR observations.

We have presented the first comprehensive data (multi-wavelength photometry and multi-epoch spectroscopy) showing an SLSN evolving into the nebular phase. The slow decline of SN 2015bn would require a radioactive ⁵⁶Ni mass much larger than the iron mass (and indeed total mass) suggested by the nebular spectrum. The spectrum does not show signs of large ⁵⁶Ni mass or strong CSM interaction. Instead, spectroscopy and light curve fitting suggests an ejected mass 7–30 M_⊙, energized by a persistent source of ionizing photons, and an inner dense region that could be a signature of overpressure from a central engine.

At ∼400 days, the spectrum is virtually identical to energetic SNe Ic such as SNe 1997dq and 2012au, and extremely similar to GRB-SN 1998bw. This suggests that the composition and inner density structure are similar, despite their very different luminosities, and confirms the long-held suspicion that SLSNe and GRB-SNe are related (Pastorello et al. 2010; Lunnan et al. 2014; Greiner et al. 2015; Nicholl et al. 2015; Mazzali et al. 2016). The similarity to GRB-SNe favors a scenario in which SLSNe result from the engine-powered explosions of stripped massive stars. Scheduled deep X-ray observations at these late epochs may reveal the engine directly.

M.N. thanks the organizers and participants of STScI workshop "The Mysterious Connection Between SLSNe and GRBs" for stimulating discussions and Carlos Contreras for providing magnitudes to check GMOS flux calibration. Based on data from ESO as part of PESSTO (188.D-3003, 191.D-0935). PS1 is supported by NASA grants NNX12AR65G and NNX14AM74G from NEO Observation Program. S.J.S. acknowledges ERC grant 291222 and STFC grants ST/I001123/1 and ST/L000709/1. A.G.Y. acknowledges ERC grant 307260. K.M. acknowledges an Ernest Rutherford Fellowship through STFC. T.W.C. acknowledges the Sofia Kovalevskaja Award to P. Schady from the Alexander von Humboldt Foundation of Germany. S.T. acknowledges support from TRR33 "The Dark universe" of the German Research Foundation.